This newsletter highlights key happenings at Lumina from the past month or two. For tips, informal demos, and some of our thoughts on effective modeling, also check out the Lumina blog.
Analytica 5.1 Release
Analytica 5.1 is now available! This release focuses on faster computation using more parallel calculation using multi-threading. It adds several helpful new features, including toggling the lines or bars on a graph by clicking on the Key — see the animation on the left. It builds on the major 5.0 release, which added a huge number of enhancements, including a more convenient user interface, powerful Find dialog, quick links to docs on the wiki, and new graphing features. See What’s new in 5.0?, and What’s new in 5.1?
Entrenamiento de Analytica en español (Training in Spanish)
Lumina está llevando a cabo un entrenamiento de Analytica de 2 días en Campbell, California, del 21 al 22 de agosto de 2018 en español. El taller de dos días (más el tercer día opcional) proporcionará una capacitación concentrada sobre cómo crear y usar modelos de decisión. Es para aquellos que tienen cierta experiencia en la creación de modelos, por ejemplo, con hojas de cálculo, que desean expandir sus habilidades de modelado. Combina la formación práctica de nuevos usuarios y aquellos con experiencia en Analytica.
Contáctenos en firstname.lastname@example.org o regístrese en nuestra página webpage.
Lumina Welcomes Lee Davidson
Lumina is delighted to welcome a new full-time software engineer – Lee Davidson – who was previously with PayPal. Lee has many years of experience as a software developer, and is working closely with CTO Lonnie Chrisman. Lee’s Ph.D. from Yale is in Philosophy, and he maintains strong interests in philosophy, logic, mathematics, and physics. Lee has written two novels and many short comic-literary writings, some published. He is also a classical pianist, with several public performances to his credit.
Analytica Tips: Does Size Matter? How to choose a Sample Size
Analytica represents and computes each uncertain quantity as a random sample from its probability distribution. You can select several uncertainty views including Probability bands, Density function, or Cumulative probability distribution. The accuracy depends on the number of samples. The default sample size of 1000 is usually fine for initial tests as you build a model. But, it may result in some noise, especially in the probability density, which is much more sensitive to random variation than the cumulative probability view. A simple way to reduce the noise is to set the Smoothing in the Probability density tab of the Uncertainty Setup dialog. We recommend the default smoothing factor. Higher levels can be misleading, e.g., with tails too wide.
Sometimes you need a larger sample size. How large? Some analysts do extensive “convergence tests” rerunning their model with various sample sizes. That’s usually a waste of time. For Monte Carlo simulation, you can figure out what sample size you need using simple statistics. Suppose you care about the mean (expected value) of a result, and want to be 95% confident that the true mean is within +-0.5% of the estimated mean. You can read the needed sample size from this graph – about 60,000. Or you can specify a desired interval on, say, 95% percentile. See Selecting the sample size for more. It includes the Choose_sample_size.ANA library which you can download with a click to let you calculate the sample size you need.